China ranks first in the world in terms of the number of published AI papers. Why hasn’t it launched ChatGPT?
Perhaps the biggest technology news in the world in recent months is the popularity of ChatGPT.
ChatGPT is an artificial intelligence application developed by OpenAI, a start-up technology company in the United States. It will be open to the public in November 2022 and has accumulated 100 million users by January this year.
I am also one of these 100 million users. My feeling about using ChatGPT during this period is that I would not rely on it if it were for economics and financial searches in my major. Since I know where the original source of most of the economic and financial data I need is, I go directly to the website where the data is published. ChatGPT may be more convenient, but since my job requires data accuracy, it is more efficient to skip the intermediate level of artificial intelligence.
But outside of professional areas, especially in completely unfamiliar areas of knowledge, ChatGPT is a very useful auxiliary tool. Just like some of the "know-it-all" friends we meet around us or on the Internet, what they say may not necessarily be the most correct authoritative opinions, but they can indeed quickly give us an answer that is easy to understand and facilitates us to further understand things.
Of course, if it were just a function like search assistance, then ChatGPT might just be a smarter Wikipedia and not worthy of the current heated attention. Its application in various scenarios other than search has become the focus of everyone's attention.
Personally, ChatGPT may be the biggest help to me in assisting in various primary tasks. At work, I often need to analyze large amounts of data to find patterns, identify trends, and study various abnormal changes. But when organizing one's own research conclusions into words, a lot of working time is actually used to complete some preliminary data analysis and text work. Before doing real deep data mining, it is often necessary to complete describing the original data and drawing pictures. Tables, simple descriptions of charts and other tedious steps.
When Microsoft announced that it would integrate ChatGPT into its own office software, in the future I can "generate" a first draft containing charts and text introductions from original economic and financial data with one click. As long as I can completely and accurately describe the original data and introduce the charts in this first draft, then I can get rid of the tedious primary labor and spend the main time on digging out the stories behind the data and looking for the real core of opportunities and risks brought about by data changes. on the task. After I complete a complete analysis, ChatGPT may even "generate" a beautiful PPT with one click, which will of course increase work efficiency exponentially.
In reality, many of us are faced with a lot of tedious text work. If ChatGPT or similar artificial intelligence software can allow people to "generate" these files with one click in the future, it will be unimaginable to improve work efficiency and the huge business opportunities derived from it. .
I believe that recently countless people are thinking about what kind of changes artificial intelligence may cause in their own fields, and many people are also asking: Why have technical achievements like ChatGPT not appeared in China?
OpenAI, the developer of ChatGPT, was founded in December 2015 as a non-profit organization and transformed into a for-profit company in 2019. Its main founders include current CEO Sam Altman and Tesla founder Musk. According to later media reports, the founders of OpenAI were themselves star giants in the American technology field, so they came forward to hire some of the top technical experts in artificial intelligence in the United States to join them, and many technical experts were able to work at the same level as themselves. Opportunities for collaboration are rare for researchers, so they would rather take a salary cut to join.
From 2016 to 2019, OpenAI’s tax records as a non-profit organization must be disclosed. According to their tax returns at that time, it can be seen that the costs and expenses are increasing every year. 2018 was the last full year of OpenAI as a non-profit organization. The total expenditure that year was more than 50 million US dollars, of which the labor cost was about 15 million US dollars. The cost of renting servers to run artificial intelligence model calculations reached 3,000 US dollars. The remaining $5 million includes office rent, travel expenses, legal fees and other miscellaneous expenses.
In 2019, OpenAI chose to transform from a non-profit organization to a for-profit company in order to absorb investment and allocate equity to employees, and received a US$1 billion investment from Microsoft. The development and release of the GPT language model that began in 2019 has also begun to accelerate. After the release of ChatGPT at the end of 2022, OpenAI CEO Sam Altman admitted that operating costs were too high and required the launch of paid services to balance the balance of payments, and Microsoft immediately increased investment. It is said that the new investment reached 10 billion US dollars.
To briefly summarize the above information:
OpenAI burned nearly $100 million in the three or four years before 2019 before launching the initial GPT model. Of course, the product at this time was still in its prototype stage and was far from the results we see today. Since it has become a for-profit company from 2019 to 2022, we do not know the cost expenditures of OpenAI, but judging from CEO Sam Altman’s speech and financing actions, the US$1 billion invested by Microsoft in 2019 may have already Most of it was spent.
With the benefit of hindsight, we can certainly say that ChatGPT, which OpenAI burned nearly $1 billion over the past seven years, is a very successful investment story. However, there were far more artificial intelligence start-ups established in the United States during the same period than OpenAI. Many similar companies that were more popular than OpenAI at the time have now disappeared, and the capital burned is of course countless.
From a financial perspective, innovative research and development in this unknown field is a long process of burning money, and the results are highly uncertain. Therefore, even star companies such as Google, Apple and Facebook have not "burned out" products like ChatGPT, let alone major Chinese Internet companies that lag behind their Western counterparts in terms of total revenue and profit margins.
Although domestic Internet companies give us the impression that they have strong financial resources, in this kind of cutting-edge R&D competition, their capital is still much weaker than that of their foreign rivals. This is actually somewhat similar to the current personal entrepreneurship. Wang Sicong once invested in entertainment companies such as Panda Live and Banana Entertainment, but later basically lost all his money. However, this does not prevent him from now being able to easily spend more than 2 million yuan for civil settlements. Most ordinary entrepreneurs do not have so many opportunities for trial and error. One failure may wipe out many years of savings, making it difficult to burn money for a long time for an illusory business goal.
When OpenAI was founded seven years ago, domestic Internet companies were still in the stage of just completing their start-up and fully developing. Alibaba has just launched China’s first “Double Eleven” shopping festival, and Tencent’s WeChat Pay is in a red envelope war with Alipay. Later, the Douyin short video platform of ByteDance, a leading domestic artificial intelligence company, had not yet been launched.
Today, stimulated by the success of ChatGPT, major domestic Internet companies have decided to follow suit and enter this field. Although this follow-up and imitation R&D strategy that focuses on certainty and reduces trial and error costs is shameful to say, it is probably the most suitable for domestic major companies at present. A rational choice based on some enterprises' own conditions.
So why have domestic scientific research institutions and university departments failed to produce similar innovative results?
If we use paper output as a standard to measure the level of the field of artificial intelligence, then China has now reached the first place in the world. According to the information provided by the "Artificial Intelligence Index Report 2022" published by Stanford University, the number of papers published by Chinese researchers in AI academic journals and academic conferences and the number of citations of these papers have reached the first place in the world, and the number of papers published is even More than twice the number published in the United States. But how much do these papers and citations contribute to economic production?
Ranking of number of papers published in AI academic journals
Ranking of citations in AI academic journals
According to the "2022 National Technology Market Statistical Annual Report" released by the Ministry of Science and Technology, all technology contracts for all majors exported by universities across the country in 2021 reached nearly 130,000, with a transaction volume of 79.04 billion yuan. In 2021, the total transaction value of all domestic technology contracts reached more than 3.7 trillion yuan, and the amount of scientific research output of universities only accounted for 2% of the value of all technology contracts in the country. As a branch research direction of artificial intelligence, the proportion of value created by its technology contracts is probably even miniscule.
Why is the proportion of scientific research output so low in places like universities where highly educated scientific researchers are concentrated?
A very important reason is that the core indicators used by domestic universities and research institutes to evaluate researchers are the number of published articles and the grade of publications. According to media reports, teachers aged 30-44 account for 59.6% of the full-time teachers in ordinary universities across the country. However, these young and middle-aged masters, doctoral students, and postdocs who have received the best education and whose scientific research abilities are in their prime period, under the pressure of the school's "promote or leave" policy, use all their energy other than teaching in the pursuit of "publishing articles." on this goal.
Of course, using the indicator of paper publication to evaluate scientific researchers is also a common practice in European and American universities. In China, it is just in line with foreign standards, and on this basis, the "test" is even more powerful.
However, we are facing a real "Iron Curtain" of science and technology today. For the country and society, scientific researchers helping companies solve several specific engineering and technical difficulties may be of greater value than publishing a few more papers in some academic journals. Some.
Xinhua News Agency calls for the end of "theory-only theory" (data map)
I have previously written an article about the example of industry-university-research cooperation in the U.S. semiconductor industry in the 1980s and 1990s. The key is that the state and enterprises jointly fund, with universities and scientific research institutes providing scientific researchers to undertake specific projects, and the enterprise's production technology backbone as the backbone. judges, to jointly solve the process and equipment difficulties faced by domestic manufacturing in key fields. If we want to leverage the group of young and middle-aged R&D personnel with the greatest research potential in domestic scientific research institutions, the state must provide corresponding policy incentives.
For example, improving the specifications of R&D projects through industry-university-research cooperation in major industries and giving these projects a status similar to National Natural Science Foundation projects, so that undertaking and completing these R&D projects through industry-university-research cooperation can help young and middle-aged teachers complete the school’s assessment indicators. Contributes to job title evaluation and promotion. This will free a large number of domestic scientific research human resources from the involution of competition papers and instead invest in practical application research to fill the shortcomings of the industrial chain.
The successful launch of ChatGPT not only shows us countless possibilities, but also reminds the country. The development of technological innovations such as artificial intelligence and quantum computers seems to be entering an era of accelerated iteration and progress. This is somewhat similar to the plot in "The Wandering Earth 2" where Tu Hengyu's daughter Tu Yaya matures rapidly through rapid iteration.
Domestic technological catch-up faces double difficulties: on the one hand, we need to catch up with the accelerating technological frontier; on the other hand, our existing shortcomings and the fact that we are blocked make it more difficult for China to catch up. For example, for the high-performance AI chips required for artificial intelligence calculations, China cannot purchase advanced foreign products, nor can it manufacture its own designed substitute products.
At this critical moment, improving investment efficiency and promoting industrial upgrading not only requires ensuring the efficiency of capital investment, but also liberating those human resources that are in "involution" so that valuable scientific research forces can be concentrated to solve the urgently needed production needs of the country. Create problems.
Recommended reading
京公网安备 11010802033920号